As the object during capture is place on the board you are never able to see the bottom of the object. During reconstruction this leads to creating a hole in the object that does not actually exist.
Like proposed there https://github.com/wg-perception/capture/issues/18 , I guess the proper way would be to flip the object and do a second capture, then use some alignment algorithm in PCL to get a fusion of the two capture and hence get a full object for reconstruction.
For now, I am worked a little on the Meshlab scripts and found a way to add a flat bottom to the object by using its convex hull. You just have to use the --close flag to add a bottom to the object.
See the difference here:
I have tested it on several objects, and it seems to work fine.
This is a fine temporary hack before we code a nice way of aligning twoo set of captures.
For example, my hack won't fix this, because there is no points below the camera lens
As the object during capture is place on the board you are never able to see the bottom of the object. During reconstruction this leads to creating a hole in the object that does not actually exist.
Like proposed there https://github.com/wg-perception/capture/issues/18 , I guess the proper way would be to flip the object and do a second capture, then use some alignment algorithm in PCL to get a fusion of the two capture and hence get a full object for reconstruction.
For now, I am worked a little on the Meshlab scripts and found a way to add a flat bottom to the object by using its convex hull. You just have to use the
--close
flag to add a bottom to the object.See the difference here:
I have tested it on several objects, and it seems to work fine. This is a fine temporary hack before we code a nice way of aligning twoo set of captures. For example, my hack won't fix this, because there is no points below the camera lens